Event History Analysis: Overview

Abstract

In event history analysis individuals are assumed to move between states. Simple cases include survival analysis with two states 'alive' and 'dead'; several types of failure or competing risks with transitions possible from 'alive' to 'dead of cause i' for i=1,...,k; and illness-death or disability models with three states - transitions are allowed back and forth between 'healthy' and 'diseased' and from each of these to 'dead'. This survey will outline the powerful counting process approach to event history analysis, and it will focus on two more recent features: the use of multistate models for prediction, both in this world and in hypothetical worlds where some transition rates are artificially changed (application: bone marrow transplantation), and the role of the sampling design and of unobserved heterogeneity ('frailty') in models for repeated events (application: repeated admissions of psychiatric patients).